Aesthetics in Digital Photography

6Primitive-based Classification Methods

The message of beauty is free of concepts.

Mikel DUFRENNE (1980)

[. . . ] once it is accepted that there is no physical or perceptual property that distinguishes works of art from simple objects, then all that remains, apparently, is proving that the difference is conceptual in nature.

Jean-Pierre COMETTI, Jacques MORIZOT, Roger POUIVET (2000)

Artistic photos are documents of large dimensions that demand that we process them in their full resolution, across their finest gradations and colors. Reducing these photos to a family of primitives is an arduous task, but it was a necessity for researchers working on the earliest machine learning methods¹. In the methods presented here, we will first look at the use of specific primitives to help in aesthetic judgment, primitives chosen to reflect the know-how deployed by experts in their evaluation of images and the criteria that we have examined at length earlier in the book (see Chapter 3). This choice justifies the generic name given to these primitives: handcrafted primitives. However, equipped with their experience in representing images for indexing and searching for data in the field of multimedia imaging, researchers also suggested that we adopt more generic characteristics, which were known to reflect properties across a large diversity of photographs: describing the light histogram, or color planes, the spatial distribution of color patches, the geometric moments of regions during ...